U.S. patent number 7,707,164 [Application Number 10/810,267] was granted by the patent office on 2010-04-27 for system and method for data cleansing.
This patent grant is currently assigned to Dun & Bradstreet, Inc.. Invention is credited to Cynthia Bergelt, Andrew Kapochunas.
United States Patent |
7,707,164 |
Kapochunas , et al. |
April 27, 2010 |
System and method for data cleansing
Abstract
A business information service provides data cleansing to
correct and update both domestic and global addresses. A
combination of processes generate cleansed data for input into a
matching process. The matching process matches information about a
business, including the address, to a unique business identifier in
at least one database of business information. The matching process
is more successful with more standard and accurate input
addresses.
Inventors: |
Kapochunas; Andrew (Oradell,
NJ), Bergelt; Cynthia (Mine Hill, NJ) |
Assignee: |
Dun & Bradstreet, Inc.
(Short Hills, NJ)
|
Family
ID: |
33131799 |
Appl.
No.: |
10/810,267 |
Filed: |
March 26, 2004 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20040225543 A1 |
Nov 11, 2004 |
|
Current U.S.
Class: |
707/999.203 |
Current CPC
Class: |
G06Q
30/02 (20130101); G06F 16/217 (20190101); G06Q
10/06375 (20130101); G06Q 30/08 (20130101); Y10S
707/99954 (20130101) |
Current International
Class: |
G06F
17/30 (20060101) |
Field of
Search: |
;707/101,102,103X,104.1,204,10,6,254,3,206,103Y ;705/1,44,400
;713/200,161 ;711/104 ;379/201.02,220.01 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
WO 01/29780 |
|
Apr 2001 |
|
WO |
|
WO 02/077769 |
|
Oct 2002 |
|
WO |
|
Other References
Search Report dated Jul. 4, 2008 corresponding to European Patent
Application No. 04758374.5. cited by other .
XP-002142880; "Information Based Indicia Program Host System
Specification"(Draft); The United States Postal Service; Oct. 9,
1996; 42 pp. cited by other .
XP-002284896; "Data Cleaning: Problems and Current Approaches";
Erhard Rahm and Hong Hai Do; University of Leipzig, Germany; Dec.
2000; 11 pp. cited by other.
|
Primary Examiner: Pardo; Thuy N
Attorney, Agent or Firm: Ohlandt, Greeley, Ruggiero &
Perle, L.L.P.
Claims
What is claimed is:
1. A method for data cleansing, comprising: receiving an input
postal address and a DUNS number; comparing said input postal
address to a standard; providing a single best postal address
derived from said input postal address, based on said comparison;
matching said DUNS number and said single best postal address to a
database having DUNS numbers associated with postal addresses, to
find a matching postal address in said database; and correcting
said input postal address based on said matching postal
address.
2. The method according to claim 1, wherein said database is an
advanced office system (AOS).
3. The method according to claim 1, further comprising: providing a
match project analysis report.
4. The method according to claim 1, further comprising: converting
said input postal address to a predetermined record layout, before
comparing said input postal address to said standard.
5. The method according to claim 1, further comprising: associating
said input postal address with a code, said code being used to
determine said single best postal address.
6. The method according to claim 1, further comprising: associating
said input postal address with a score, said score being used to
determine said single best postal address.
7. The method according to claim 1, wherein said standard is
selected from the group consisting of: ZIP+4 coding, coding
accuracy support system (CASS), Locatable Address Conversion System
(LACS), delivery sequence file (DSF), and National Change of
Address (NCOA).
8. The method according to claim 1, further comprising: sending a
status notification to a user who supplied said input postal
address.
9. A computer implemented system for data cleansing, comprising: a
pre-auditor that generates a report having a plurality of views of
an input address file, said input address file including a record
having an input postal address and a DUNS number; a verifier that
finds and removing any invalid records from said input address
file; a vendor interface that sends said input address file and an
order to a vendor, and receives a file of standardized postal
addresses from said vendor; a component that compares said input
postal address to said standardized postal addresses; a component
that provides a single best postal address derived from said input
postal address, based on said comparison; a matcher that matches
said DUNS number and said single best postal address to a database
having DUNS numbers associated with postal addresses, to find a
matching postal address in said database; and a component that
corrects said input postal address based on said matching postal
address.
10. The system according to claim 9, further comprising: an
investigator that investigates any address not matched, upon
request.
11. The system according to claim 9, wherein said plurality of
views includes a view selected from the group consisting of:
alphabetical, most frequent content, and alpha characters only.
12. The system according to claim 9, wherein said vendor
standardizes postal addresses using at a process selected from the
group consisting of: Locatable Address Conversion System (LACS),
delivery sequence file (DSF), and National Change of Address
(NCOA).
13. A machine readable medium having instructions stored thereon to
perform a method for data cleansing, comprising: receiving an input
postal address and a DUNS number; comparing said input postal
address to a standard; providing a single best postal address
derived from said input postal address based on said comparison;
matching said DUNS number and said single best postal address to a
database having DUNS numbers associated with postal addresses, to
find a matching postal address in said database; and correcting
said input postal address based on said matching postal address.
Description
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to providing a business information
service, and more particularly, to cleansing data associated with
customer lists.
2. Description of the Related Art
Some potential business information service users have customer
data that is not functioning at the maximum possible efficiency.
This is because some critical data is missing, some addresses are
wrong, and some of the customers have moved. These problems can
affect internal databases preventing accurate identification of a
customer coming in from a telecenter, mailroom, or website, leading
to a creation of duplicates and possible mishandling the customer
relationship. Response rates to mailed promotions may weaken as
fewer customers actually receive them. There is a need for a
business information service that cleanses data to provide accurate
customer addresses.
Some services provide a mish-mash of many, often conflicting
suggested changes for each address element. This makes leveraging
corrections very difficult. There is a need for an output of a
single best correction for each address element.
BRIEF SUMMARY OF THE INVENTION
The present invention is directed to a system and method for data
cleansing that meets these and other needs.
One aspect is a method for data cleansing. At least one input
address is received. The input address is compared to at least one
standard and a single best address corresponding to the input
address is provided based on the comparison. In some embodiments,
the single best address is matched to a database having unique
business identifiers associated with addresses to find a matching
address, which is provided. In some embodiments, the database is an
advanced office system (AOS). In some embodiments, a match project
analysis report is provided. In some embodiments, the input address
is converted to a predetermined record layout, before comparing it
to the standard. In some embodiments, the input address is
associated with at least one code that is used to determine the
single best address. In some embodiments, the input address is
associated with at least one score that is used to determine the
single best address. In some embodiments, the standard is at least
one of the following: ZIP+4 coding, coding accuracy support system
(CASS), Locatable Address Conversion System (LACS), delivery
sequence file (DSF), and National Change of Address (NCOA). In some
embodiments, a report is provided. In some embodiments, the report
is a postal summary report or a pre-audit report. In some
embodiments, at least one status notification is sent to the user,
who supplied the input address.
Another aspect is a system for data cleansing comprising a
pre-auditor, a verifier, a vendor interface, and a user interface.
The pre-auditor is for generating a report having a number of views
of an input address file, which contains a plurality of addresses.
The verifier is for finding and removing any invalid records from
the input address file. The vendor interface is for sending the
input address file and an order to at least one vendor and for
receiving an output file from the vendor(s). The user interface is
for providing a single best address for each address in the input
address file. In some embodiments, the system includes a matcher
for attempting to match any address in the output file or the
invalid records to a matching address in a database that contains
unique business identifiers associated with addresses. In some
embodiments, the system includes an investigator for investigating
any address not matched, upon request. In some embodiments, the
pre-auditor calculates a plurality of counts associated with the
input address file. In some embodiments, the input address file
includes a plurality of records and each record includes a
plurality of fields. In some embodiments, the counts are at least
one of the following: a number of distinct values by field, a
missing field count, a total number of records, or a percent of
distinct values. In some embodiments, the views are one of the
following: alphabetical, most frequent content, and alpha
characters only. In some embodiments, the vendor standardizes
addresses using one of the following: Locatable Address Conversion
System (LACS), delivery sequence file (DSF), and National Change of
Address (NCOA).
Another aspect is a machine readable medium having instructions
stored thereon to perform a method for data cleansing. A machine
readable medium is any storage medium, such as a compact disk (CD).
At least one input address is received. The input address is
compared to at least one standard and a single best address
corresponding to the input address is provided based on that
comparison. In some embodiments, the single best address is matched
to a database having unique business identifiers associated with
addresses to find a matching address and a matching address is
provided.
These and other features, aspects, and advantages of the present
invention will become better understood with reference to the
drawings, description, and claims.
BRIEF DESCRIPTION OF THE DRAWINGS
FIGS. 1A and 1B are logic flow diagrams of an example method of
data cleansing;
FIG. 2 is a logic flow diagram of another example method for data
cleansing;
FIG. 3 is a logic flow diagram of the operation of an example
system for data cleansing;
FIG. 4 is a logic flow diagram of an example vendor domestic
address cleansing system; and
FIG. 5 is a logic flow diagram of an example vendor international
hygiene system.
DETAILED DESCRIPTION OF THE INVENTION
FIGS. 1A and 1B show an example method of data cleansing. In step
100, a project manager receives a user input file and file layout
and uploads the file to a processor, such as a mainframe. In step
102, the project manager sends an order with a product code to a
vendor. In step 104, the project manager sends the order form and
other information to a gatekeeper. In step 106, a pre-audit is
performed. If there is no critical error discovered by the
pre-audit, then in step 108 the gatekeeper sends a pre-audit report
to the project manager. In step 110, the project manager reviews
the report with the user and others. In step 106, if there is an
error discovered by the pre-audit, then in step 112, the process is
halted to determine if processing is to continue. If the process is
halted, then in step 114, a standard input layout for file transfer
is created. If the process is not halted, then in step 116, the
file is returned to the user. In step 118, files are split for
vendors into domestic records 120 and foreign records 122, which
are processed separately. In step 123, files and an order form are
sent to a vendor, who verifies receipt of them. In step 124, files
returned from the vendor are received. In step 126, when files are
returned for foreign records, the project manager receives postal
reports from the gatekeeper and prepares a postal summary report.
In step 128, domestic and foreign files are merged into one file
with a standard layout for processing. In step 130, files are
processed and a technician sends the project manager files for
analysis. In step 132, an analysis file is created and in step 134,
the project manager send the analysis to the user.
FIG. 2 shows an example method for data cleansing. In step 200, a
qualifying field audit is performed. In step 202, addresses are
standardized, corrected, and ZIP+4-coded. In step 204, addresses
are additionally corrected, and marketing-oriented information is
appended. In steps 206 and 208, addresses are updated with changed
information, when appropriate. In step 210, new addresses are
re-processed to verify corrections and add categorization data. In
step 212, output is edited to a single best address for each parsed
data element along with selected postal codes and the original
address. In step 214, the best address is matched to a business
information database and, based on appended codes, additional
corrections are made available. In step 216, a layout data
dictionary with suggestions for leveraging postal data is
generated. In general, the example method includes processing
domestic addresses including data discovery, postal pre-processing,
and, optionally, matching.
Data discovery begins with the pre-audit and includes parsing and
reformatting a customer file and verifying that a large number,
such as 85% of the records in the customer file have enough address
elements to be helped by postal pre-processing. It is verified that
there is one address per record. Variations of an address on a
single record, i.e., a bill-to and a ship-to, or a street address
and a P.O. Box, need to be "exploded" into separate records to be
helped by postal processing. It is verified that the data is for
the United States only. Different processes are used for foreign
data. The pre-audit also includes examining the contents of every
field in every record, and a report is produced, which applies
letter grades to each data element, reflecting completeness and
relevance.
In step 202, postal pre-processing is performed through a
combination of processes and matching to multiple USPS-compiled
database, such as a database totaling over 280 million domestic
records, for corrections. Standardization, correction, and ZIP+4
coding (a/k/a Coding Accuracy Support System, CASS.TM., processing)
are performed for all domestic addresses, business or consumer.
In step 204, postal pre-processing in this method also includes
applying a file to correct records and append codes, such as "good
address, but vacant for the last 90 days" and score each record for
accuracy and deliverability. One example file is a second
generation delivery sequence file (DSF2). The DSF2 is a file
containing substantially all valid addresses serviced by the Postal
Service. This comprehensive system enables the substantial
elimination of undeliverable addresses, allows mailers to obtain
additional postage discounts, and provides valuable information
about the make-up of addresses on files. The DSF2 is updated
monthly with transactions supplied by the USPS and has 156 million
address records for nearly every deliverable address in the United
States.
In step 206, postal pre-processing also includes utilizing address
standardization and DSF2 corrections to match to another file, such
as the Locatable Address Conversion System (LACS) file. LACS is a
file made available by the United States Postal Service (USPS) that
provides access to new, changed addresses for locations that have
not moved. The LACS has about 5 million records. The vendor
receives monthly updates to the USPS LACS file.
Using data that has already been standardized and corrected
increases the match rate to the LACS file. The LACS file has
addresses changed by the United States Postal Service (USPS) either
when a community chooses to provide 911 service, which requires a
building number and street address rather than a rural route box
location, or when a street name has been changed.
In step 208, postal pre-processing also includes utilization of
corrected and updated addresses from the preceding steps to match
to another file, such as the weekly updated 120-million-record
National Change of Address (NCOA) file.
The NCOA file is made available by the USPS to provide mailers
current change of address information so as to reduce undeliverable
mail and increase response rates. This comprehensive system
identifies and corrects addressing errors before mail enters the
mail stream. A vendor receives updates to the NCOA file every week.
NCOA covers four years of moves, with additional possible moves (on
near matches to a "from" address) flagged via NCOA-Nixie footnotes.
The NCOA has about 120 million records in a rolling four-year
database of from- and to-addresses, requiring an almost perfect
match to the old name and address to get a new address appended.
The NCOA-Nixie flags include a reason code why a new address could
not be appended.
In step 210, new addresses generated from NCOA are then
reprocessed: first against LACS and then against DSF2. New
addresses coming from LACS that were also not NCOA matches are
reprocessed against DSF2.
In step 212, postal pre-processing results in a set of best address
corrections or address updates for each address element. The best
address corrections or address updates are appended to the input
address, avoiding the creation of a file with multiple and
conflicting sets of corrections for each address element as is the
common practice from conventional processes.
In step 214, the results are matched to another file, such as a
31-million-record advanced office system (AOS) file. A certain
number of postal processed records have either failed to be
recognized by postal processing, or failed to be completely
corrected. For instance, records with missing or wrong suite
numbers. Historically, matches, at some level of confidence, are
made for 30% to 95% of the records that postal processing
determines to be uncorrectable. If such a record is matched to a
database, (allowing for a lower confidence match is normally
acceptable, because it is already known that the client address is
incorrect) and if the user agrees the match is valid, the user has
the option to further correct the record by using address elements
from the matched record in the database.
An example method of data cleansing provides address correcting and
updating service for domestic and global address records using a
combination of processes. The domestic method includes the
following steps: (1) in step 200, performing a qualifying field
audit; (2) in step 202, standardizing, correcting, and ZIP+4 coding
address records via CASS-certified software; (3) in step 204,
correcting and appending marketing information via DSF; (4) in step
206, updating the address records via USPS LACS; (5) in step 208,
updating the address records via USPS NCOA and NCOA-Nixie flagging
of possible moves; (6) in step 210, applying NCOA for new addresses
from LACS, and applying DSF to NCOA addresses, to make certain all
addresses have maximum corrections and appended data; (7) in step
212, editing output to a single best address for each parsed
address element, along with selected postal codes, and the address
as originally submitted; (8) in step 214, matching the best address
to a domestic business database, and, based on appended codes,
making additional corrections on records that match to the
database; and (9) in step 216, providing a layout or data
dictionary with suggestions for leveraging postal data. A project
manager initiates a field by field audit and a multi-step
standardization, correction, and updating process, preferably in
three days or less.
Data cleansing includes applying a decision tree to derive a
domestic best address. The highest priority is addresses with a
positive match to the NCOA file. NCOA-generated addresses are
re-processed through address standardization, DSF, and LACS to
ensure validity, but are still called NCOA addresses and have an
appended move date. An NCOA address, when it is a brand new street,
for instance, can be a street name not yet on the DSF file. In such
cases the NCOA address stands and is delivered as the best address.
The next priority is new addresses gained through LACS that do not
match to NCOA. Addresses would be DSF processed on a second pass to
validate. The next priority is addresses cleansed through DSF that
do not match NCOA or LACS. The next priority is addresses that
match address standardization, but not DSF. The last priority is
addresses failing to match address standardization. These addresses
are parsed and are used to populate the best address fields.
Data cleansing for foreign addresses includes a project manager
initiating an audit and then reformatting, correcting,
standardizing and appending a single set of best addresses to an
original record or records. Preferably, software containing the
best available global postal agency information is used.
The global method includes the following steps: (1) performing a
qualifying field audit; (2) parsing, reformatting, and correcting
city, state/county/prefecture and country names and properly
formatting postal codes; (3) applying global postal standardization
and correction software; (4) coding output records; (5) appending a
single best address for each parsed address element to the address
as originally submitted; (6) matching the best address to at least
one business database, and, based on appended codes, optionally
making additional corrections on records that match to the records
in the database. An example of record coding for step (4) is: valid
as submitted, corrected, valid after corrections, possibly
deliverable; not standardizable or correctable, but appears to have
all required address elements for a specific country, possibly
because that country does not provide address information that
would enable verification/correction, or probably undeliverable,
either because two or more critical address elements are missing or
because the address has an uncorrectable, pre-unification, German
postal code.
Another example method for data cleansing includes receiving a
file, such as a flat file on a CD, cartridge, email, etc. An audit
is performed on the file to verify that name and address fields are
adequately populated. If so, domestic or global processing is
performed for postal processing and address correction and
standardization. Preferably, the domestic or global processing is
performed by a vendor. The result is one best address for a given
input address. Then, the best address is matched to a database of
business information.
FIG. 3 shows the operation of an example system for data cleansing.
In step 300, the program manager documents user requirements. In
steps 302 and 304, profiles are created based on user-defined
requirements. In step 306, a user input file is received. In step
308, a pre-audit is performed. In step 310, a pre-audit report is
generated and made available to others, such as by posting to a
website. In step 312, the program manager reviews and sends the
report to the user. In step 314, invalid records are separated and
put into a separate file, which will be appended to the valid file
received from a vendor in step 328. In step 316, an order form and
other information is sent to the vendor in a separate file, ahead
of the data file. In step 318, the vendor processes the
information. In step 320, a postal summary report is generated by
the vendor and received by the program manager. In step 322, the
program manager reviews the results, creates a summary presentation
and shares them with others. In step 324, the user reviews the
results. In step 326, the file is received from the vendor. In step
328, the invalid record file (from step 314) is combined with the
returned vendor file. In step 330, matching and appending is
performed. In step 332, a results report is generated and made
available to others. In step 334, the program manager generates a
project analysis report. In step 336, the program manager reviews
the results and sends them to the user. In step 338, it is
determined whether an investigation is requested for unmatched
records. If so, in step 340, the unmatched records are processed.
In step 342, additional results are made available to the user. In
step 344, the user receives results as they become available.
In general, the example system receives user input addresses,
processes them, and provides a file having updated addresses, a
postal processing summary report, a match project analysis report,
and a pre-audit report. The system is preferably capable of
handling about 250,000 records sent monthly by about 400 users.
Preferably, the system provides output in 72 hours or less for
domestic addresses and 10 days or less for foreign addresses. The
system tracks the status of processed data throughout the process.
The system sends notifications to the user, e.g., email messages,
at various points in the process, such as upon receipt of an input
file or when an error occurs. These notification emails are sent to
internal and external customers, whenever there is activity on
accounts that they are monitoring. Input files may be in any format
and may be encrypted or compressed. The system provides a
recommended but not required layout to the user. Preferably, users
separate domestic and global addresses. Input files may include
unique business identifiers, such as DUNS numbers, that correspond
to identifiers in the matching databases. An input file is
transmittable to the system through the Internet or a leased line.
Preferably, batch processes are used to transfer input files.
When the user attempts to login to the system, they are prompted
for a user ID and password. Successful login brings the customer to
the root of their directory structure. From the root directory the
customer has an option to change directories to their puts (deposit
files), or their gets (retrieve files) directory.
The example system decompresses the file, if it has been
compressed, decrypts the file, if it has been encrypted with PGP,
and scans the file for viruses. Then the system sends a file
accepted email to the user. The system then pushes the file to an
appropriate downstream application and sends a notification of new
request email (e.g., file has been submitted) to the user. A
downstream application is an internal application to which an
inbound file is dispatched, or the internal application from which
outbound file processing originates. A viewable status file is
selectable by the user. A process to automate file retrieval is
also available to the user. Example status files include a
filename, profile ID, tracking ID and status code and the like.
The input file is processed to have a predefined record layout,
such as the one shown in Table 1 below.
TABLE-US-00001 TABLE 1 Example record layout Start End Length
ContactFirstName 1 20 20 ContactMiddleName 21 40 20 ContactLastName
41 60 20 AddressLine1 61 124 64 AddressLine2 125 188 64
AddressLine3 189 252 64 AddressLine4 253 316 64 City 317 380 64
State 381 400 20 PostalCode 401 410 10 CountryName 411 430 20
Business Name 431 550 120 Phone # 551 565 15 DUNS # 566 574 9
Filler 575 584 10 Our Sequence # 585 591 7 Our Sub-sequence # 592
592 1 I' Indicator 593 593 1
The example system includes a pre-auditor, verifies various aspects
of the input addresses, and calculates frequency counts for various
fields in the records, such as company name, address 1, address 2,
address 3, address 4, city, state, ZIP and country name. The
pre-auditor calculates a number of times one of these fields is
repeated, and absence counts, presence counts, number of records
and the percentage distinct within each field.
The pre-auditor generates a report including various views of the
data, such as all counts, as alphabetical, most frequent content,
or alpha characters only.
The pre-auditor generates an all-counts view of the data. For each
field in the records, counts are calculated, such as a number of
distinct values by field for all records (# of unique values by
field), an absence count (number of records missing content for
specified field), presence count (number of records populated with
content for specified field), number of records (total number of
records in the file), percent distinct (percent of distinct values
compared to total of records in file (percent=number of distinct
values/number of records in the file). The total number of records
also equals the total of absence and presence counts. For example,
examining the company name field for a file yields the following:
the file contains 1,000 records for the company field, 850 records
are distinct values, 100 records are absent, and 900 records are
present.
The pre-auditor generates an alphabetical view of the data. For
each field in the record, the pre-auditor shows a predetermined
number, such as 50, of the first occurrences of information within
the field sorted alphabetically, preferably in ascending order. For
each unique field content, the pre-auditor determines a number
count of duplicates, displays the first predetermined number of
occurrences by occurrence name, determines the number of
duplicates, determines the percentage of occurrences compared to a
total number of records in the input file, and determines a number
of occurrences for particular fields per the number of total
records in the input file. An example is shown in Table 2
below.
TABLE-US-00002 TABLE 2 Alphabetical view Specified Field Count
Percentage of file that has (i.e. Company Name) (Occurrences)
occurrence Sort alphabetically in How many times the Percentage of
occurrences ascending order. (Company Name) compared to total # of
Content of specified occurs in the file records in file (% = # of
field occurrences/# of total records in file) Example: Example:
Example: A&A Investment 3 (Company Name 0.01% (Company name
Network Inc DBA occurs three times makes up 0.01% of file) Sub in
file)
The pre-auditor generates most frequent content view of the data.
For each field in the input records, a predetermined number, such
as 50, of the highest frequencies or occurrences within the field
is determined. For each unique field content, the pre-auditor
determines a number of duplicates and displays the first
predetermined number of occurrences of most repetitive field
content that occurs in the file, giving occurrence name, number of
duplicates, and percent of occurrences compared to the total number
of records in the file. An example is shown in Table 3 below.
TABLE-US-00003 TABLE 3 Most frequent content view Specified Field
(i.e. Company Percentage of file that Name) Count hasoccurrence
Content of Sorted in descending order Percentage of specified field
according to the highest occurrences compared to (i.e. Company
occurrence on the file, how total # of records Names) many times
does the in file (% = # of (Company Name) occur in occurrences/# of
the file records in file) Example: Example: Example: Edward A
Kaplan 40 (Occurs 40 times in file) 0.12% (This company DBA Edward
A name makes up 0.12% Kaplan of file)
The pre-auditor generates an alpha characters only view of the
data. For each of the fields, the pre-auditor displays a
predetermined number, such as 50, of the highest frequencies or
occurrences of records containing non-numeric, alpha-numeric
characters within a specified field (i.e., A-Z, 1-9 and a blank
space). Unacceptable occurrences include more than 1 occurrence of
anything other than alpha-numeric characters. For each unique
field, content with alphas only includes a count of the number of
duplicates, the first predetermined number of occurrences, the
occurrence name, the number of duplicates, and the percent of
occurrences compared to total number of records in the file. An
example is shown in Table 4 below.
TABLE-US-00004 TABLE 4 Alpha characters only view Specified Field
(i.e. Company Percentage of file that has Name) Count occurrence
Content of Sorted in descending order Percentage of occurrences
specified field according to the highest compared to total # of
(Company occurrence of special or records in file (% = # of Name)
non-printable characters in occurrences/# of records in the file,
how many times file) does the (Company Name) occur in the file
Example: Edward A 40 (Occurs 40 times in file) 42.39% (This company
Kaplan DBA name makes up 42.39% of Edward A file) Kaplan
The example system removes any invalid records from the input file
and stores them in a new file. An invalid indicator with
indicators, such as "I" for invalid or "V" for valid are added to
the record. This file is not processed, but rather held until the
rest of the input file is processed and then combined with results
file and sent to a matching process.
There are various rules for determining invalid records. For
example, for domestic records, valid combinations include: address
1 and city and state, address 1 and ZIP, address 2 and city and
state, address 2 and ZIP, address 3 and city and state, address 3
and ZIP, address 4 and city and state, address 4 and ZIP. If no
street address is present, address.sub.--1, address.sub.--2,
address.sub.--3, and address.sub.--4 are checked. If addresses 1,
2, 3 and 4 are blank, the record is ineligible. The record is
ineligible if address.sub.--1, address.sub.--2, address.sub.--3 or
address.sub.--4 is present, but there is no ZIP code or city/state
combination. For domestic records, invalid combinations include: no
address present, address 1 and city (no ZIP, no state), address 2
and city (no ZIP, no state), address 3 and city (no ZIP, no state),
address 4 and city (no ZIP, no state), address 1 and state (no ZIP,
no city), address 2 and state (no ZIP, no city), address 3 and
state (no ZIP, no city), and address 4 and state (no ZIP, no
city).
The example system includes a vendor order form processor. In an
example manual process, a program manager completes an order form
for each input file. In an example automated system, the
information on the order form is provided to a technician, who
verifies the information. This information is sent to a vendor in a
control file and is received prior to the data file. Both vendors
use the same control file layout. This information is also used to
send a vendor postal summary report to the program and to generate
a bill for files processed.
The example system includes an example user interface including a
template of the information sent to the vendors. The program
manager and customer define profile needs and order form
information. A profile is a set of characteristics and
specifications for customer file transfers as defined by
administrator entries into the user's account through an
administrative interface. An administrative interface is a user
interface for accessing a system for viewing, monitoring, and
managing user accounts and profiles. The order form is
automatically captured and electronically communicated to the
vendors. An example order form is shown in Table 5 below.
TABLE-US-00005 TABLE 5 Example order form Re- Read Field Name
quired? Only? Source Contract ID (free form) Y Program Manager Our
Contact Name Y Program Manager Our Phone Y Program Manager Our
Email Y Program Manager File Quantity Y Calculated (based on
initial number of records from BDE) Multiple File indicator Y
Program Manager Vendor Needs: (Only Y Defaults are: DSI will be
using this Maintain Diacritics = No data but it will appear Reject
USA Records = Yes on Axiom's) Canadian NCOA = No Maintain
Diacritics Reject USA Records Canadian NCOA
The example system includes a file transfer protocol (FTP) program.
Files are sent to the vendor upon receipt. Preferably, the files
arrive individually in order for the vendors to process the post
summary report for each job and send the post summary report to the
program manager. Bundling multiple files is also an option.
The example system including completing the pre-audit, creation of
a control file, and creation of an input file for each vendor. An
example layout of the input file is shown in Table 6 below.
TABLE-US-00006 TABLE 6 Example layout of input file Start End
Length ContactFirstName 1 20 20 ContactMiddleName 21 40 20
ContactLastName 41 60 20 AddressLine1 61 124 64 AddressLine2 125
188 64 AddressLine3 189 252 64 AddressLine4 253 316 64 City 317 380
64 State 381 400 20 PostalCode 401 410 10 CountryName 411 430 20
Business Name 431 550 120 Phone # 551 565 15 DUNS # 566 574 9
Filler 575 584 10 OurSequence # 585 591 7 Our Sub-sequence # 592
592 1 I' Indicator 593 593 1
The example system includes a vendor output file receiver. The
output file receiver sends a notification of receipt.
The example system includes a vendor-to-user linker. An incoming
file from a user is linked to a vendor. When an output file is
received from the vendor, the linker returns the output file to the
user. Vendor files are combined with the invalid record file from
the pre-audit process. This file includes raw user input data and
postal pre-processed data or the user data and no postal
pre-processed data for invalid records. The valid and invalid
records are combined and a single file is sent to the matcher.
The example system includes a matcher. The following fields are
mapped: the original company name from the user, address from the
vendors, and original phone number from the user. If the addresses
are blank, then the original user address is used. If address
information from a vendor is blank, then the matcher matches
against the original customer address information.
The example system includes a project creator. A match technician
creates a new project, renames an output file and uses new or
original customer address information to perform matching. Users
send a second file using a different profile in a batch file. A
file is received from a vendor and matching is performed per
profile instructions. Resulting matched records are sent to an
appended file in the example system and unmatched records are sent
to an investigator in the example system, if requested by the
user.
The example system includes external interfaces. Files are sent and
received from vendors. The system sends the original customer
address to a vendor. The vendor sends the best corrected address
back along with the original customer address and postal code
information. Preferably, standard input and output layouts are
used.
FIG. 4 shows an example vendor domestic address cleansing system
that standardizes addresses according to USPS specifications. In
step 400, a source file is posted to an FTP site 402, address
cleansing is performed 404, DSF and LACS processing is performed
406, and NCOA processing is performed 408, and addresses are
reformatted and components are selected 410.
The system enhances the user's data by verifying and correcting
5-digit ZIP codes, applying ZIP+4, delivery point barcodes, carrier
route codes, and line of travel data. The system also ensures a
CASS-certified output. CASS is the USPS certification process for
address standardization products, which is updated and re-certified
annually.
The vendor address cleansing system has a reformat address
component selection. This component reformats output records to
comply with the standard output layout. The process also ensures
that the optimum address components are selected from DSF/LACS/NCOA
based on priorities set by the vendor.
FIG. 5 shows an example vendor international hygiene system. In
step 500, conversion is performed to review data, correct initial
problems, and correct problems discovered in a first pass of phase
one. In step 502, phase one is performed, including country
isolation and name standardization, postal code isolation and
reformatting, state or province isolation, review of rejects and
possibly rerun the conversion. In step 504, filters are applied for
obscenity detection and miscellaneous garbage detection. In step
506, domestic records are split off. In step 508, phase two is
performed, including postal code validation and correction, city
validation and correction, and street validation and correction,
where available. Instep 510, Canadian NCOA is performed, if
requested.
The present invention has many advantages. For first class mailers,
the user's mail, such as invoices, is forwarded to new addresses
when the addressees move, but having the new address in advance
saves one to two weeks of delivery time. For standard class (bulk)
promotions, more pieces are delivered with more accurate addresses
yielding a higher response rate. For all businesses, data cleansing
facilitates internal data integration efforts and generates high
match rates to other data. Cost savings are realized, depending on
the size of the customer list. The present invention is able to
determine a correct address and match it to a unique business
identifier in a database for up to 95% of the addresses determined
to be uncorrectable by the U.S. Postal Service. The present
invention has a database with nearly 19 million marketable U.S.
business records and 14 million more in an historical repository.
The present invention appends data that is about 98% ZIP+4-coded
due to monthly address updating and maintenance routines. For
international addresses there are about 41 million marketable
records. The matcher may provide an improved address even when
postal correction software is unable to.
It is to be understood that the above description is intended to be
illustrative and not restrictive. Many other embodiments will be
apparent to those of skill in the art upon reviewing the above
description, including other systems and methods for data cleansing
and other similar differences. The present invention applies to
many fields where data is cleansed. Therefore, the scope of the
present invention should be determined with reference to the
appended claims, along with the full scope of equivalents to which
such claims are entitled.
* * * * *